digital sculpting

8/8/2019 Digital Sculpting

1/13

COPYRIGHT 2000, NICHIMEN GRAPHICS

DIGITAL SCULPTURE TECHNIQUES

HOWTO APPLYTHE PRINCIPLESOF TRADITIONAL SCULPTURE

TO MAKE STUNNING 3D CHARACTERS

BY

BAY RAITTAND GREG M INTER

All images by Bay Raitt


2/13

2DIGITAL SCULPTURETECHNIQUES


Even among technically oriented artists, 3D is perceived to be a left-brain medium And, in large de-

gree, it is. The mathematical rigors of representing 3D objects, complete with environmental effects

such as light, continue to leave their mark on even the most user-friendly modeling and animation

tools. Newcomers have to learn a new way of thinking while even the most experienced 3D artistscontinue to find new uses for abstruse features. Once you step into the third dimension, you leave

the familiar world behind.

Searching for the keys to more productive 3D, nearly everyone misses an obvious, quite familiar,

extremely relevant discipline: sculpture. Considering 3D modeling as a process of sculpting sug-

gests powerful techniques that can simplify difficult 3D tasks, especially the task of modeling char-

acters for animation.

Figure 1 When a sc ulptor c hisels a form from a bloc k of ma rble, the bloc k bec ome s sma ller. Similarly, a co ntrolobjec t g enerates a sma ller lderived surfac e.

In this article, well look at ways to manage the complexity of even the most elaborate character an-

imation using the techniques of volume modeling, derived surfaces, and edge loops. The workflow

described mimics the traditional sculpting process and lets you generate output in a variety of reso-

lutions for VRML, games, animation, film, or print from a single source. This approach is an excel-

lent alternative to the shortcomings of NURBS modeling and its well on its way to becoming a

primary paradigm for 3D production. In fact, Pixar has started to reorient their entire production

pipeline to accommodate it.

At the moment, Nichimens Mirai and Kinetix 3D Studio Max are the only tools that fully support

this approach (MAXs implementation of derived surfaces is only partial). NewTek LightWaves

metaNURBS capability does the trick for high-resolution output only.


3/13



VOLUME MODELING

A sculptor starts by rapidly defining the volume, then concentrates on refining the surface. In 3D,

this method is known as volume modeling, the process of growing, extruding, and manipulatingedges, faces and vertices of a polygonal primitive such as a cube. In contrast, when you work with a

modeler that manipulates higher-order surfaces, including NURBS, patches, and metaballs, you

start by defining an objects surface first, and then modifying its volume. To a traditional sculptor,

this approach puts the cart before the horse.

Sculpting in the digital realm lets the sculptor work naturally, but with a superset of the tools avail-

able to the traditional artist. He can twist a model around its axis, duplicate its geometry around a

face, even squash and stretch it with impunity, secure in the knowledge that any changes can be un-

done. The ability to save work at different stages or undo multiple topological changes is beyond

the realm of possibility when youre working with clay, stone, or other physical media. A sculptor

working on a clay model to be digitized has no such freedom.

Moreover, because youre working with a relatively simple mesh rather than higher-order surfaces,

you can do most of the work in real time. The traditional knock against polygonal modelers has

been that machines bog down under the load of a high-polygon, textured model. This was true a

year or two ago, but today you can pick up a fully loaded 400 MHz dual-Pentium machines with

screaming 3D board for a couple grand. A machine like that will move even high resolution models

around at head spinning speed.

The creative benefits of speed and simplicity may seem obvious, but many 3D artists work with

meshes so complex that only parts of it can be seen on the screen at any given time. In this situation,

you forfeit the spontaneity essential to creating rich, expressive forms. Volume modeling lets you

focus on building an efficient low resolution form while taking advantage of the power of your soft-

ware and hardware to generate higher resolution versions using derived surfaces.

DERIVED SURFACES

A typical characters low resolution form might be 500-1000 polygons, depending on its structure.

While a character of this resolution is probably acceptable for many real time applications, its defi-

nitely unacceptable for production animation, film, or print. Fortunately, most of the current gener-

ation of polygonal modelers can generate a separate higher-resolution mesh automatically from a

lower resolution version via smoothing. The higher resolution version is referred to as a derived sur-

face.

Using Mirai, MAX, or LightWave, the original low resolution object can be used as a control object

to manipulate the derived surface. If your software takes care of everything it should (such as main-

taining texture integrity on subdivided surfaces), youll never have to manipulate the derived surface

directly.

When a sculptor chisels a form from a block of marble, the block becomes smaller in dimension.

The control object serves as a bounding volume for the derived surface, simulating this fundamental

principle of real-world sculpting (Fig. 1). The control object can be smoothed once, twice, or as

many times as necessary, depending on the output resolution you require. Figure 2 shows the evolu-

tion of a character from an artists sketch to a high-res version. The figure in the middle is the con-


4/13



trol object, the one to its right is the derived surface, and the figure to the far right (also a derived

surface) has been smoothed twice and rendered.

Figure 2 The e volution o f a c harac ter. The a rtists sketc h is at far left. To its right is a q uick low -resmod el that has been te xtured for reference. The figure in the midd le is the finished co ntrol object,the one to its right is the d erived surfac e, and the figure to the far right ha s bee n smoo thed twiceand rendered.

MODELINGFOR SMOOTHING

The trick is to build a control object that generates the derived surface you have in mind. To accom-

plish this, its necessary to understand the smoothing algorithm of your modeler.

Consider smoothing a simple cube (Fig. 3). A higher resolution mesh is derived from cube, while

the cube itself is unaffected. Scaling a face on the control object simultaneously updates the derived

surface (Fig. 4). Small topological changes, such as adding edges around the middle of the cube can

create a very different high resolution mesh (Fig. 5); the additional points on the control object will

let you manipulate the shape of the object further. Maintaining sharp edges on the derived surface is

achieved by pinching edges closer together on the control object (Fig. 6).

Figure 3 A low-resolution o bjec t g ener-at es a highe r-resolution d erived surfac e.In Mirai, youd create a c ube, select t hebody o f the object in the body ed it menu,ch oose Smoo th, and selec t the With His-tory opt ion.*

Figure 4 Sca ling a fac e on the cubesimultane ously upd at es the de rived sur-face.


5/13



* To achieve the same results in Kinetix 3D Studio Max, youd create a reference object and add a smoothing

operation to its stack, then shade the reference object and make it insensitive. In NewTek LightWave 3D,

create a cube and hit TAB, which replaces the cube with a smoothed metaNURBS object.he vertices of the

cube are still visible and available for manipulation if you hit TAB again.

A basic principle of smoothingand a very important oneis that although the derived surface is

smaller than the control object, it intersects the control object at the midpoint of each of the control

object's faces. These points are the key to predicting the results of smoothing. If you need a sharp

edge, create a face with a midpoint at the intended edge and then add small faces around it (Fig. 7).

Figure 7 If you need a sharp edge, create a face with a midpoint at the intended edg e and thenadd sma ll fac es around it.

Mirai lets you remove unwanted topology from the derived surface without disrupting the relation-

ship between the control object and the derived surface. You can do this by using Dissolve on seg-

Figure 5 Sma ll topologica l cha nges, such

as ad ding edg es around the middle ofthe cub e, can create a very differenthigh-resolution me sh.

Figure 6 Maintaining sharp edges in the

derived surfac e is ac hieved b y pinchingedg es closer togethe r on the controlobject.


6/13



ments and Merge Faces on collections of faces. (If left unreduced, the generated mesh has roughly

four times the polygons of the original control object.)

Figure 8 Mirai lets you reduce polygons on the derived surface without disrupting the relationshipbetw een the c ontrol object and the d erived surfac e. This lets you opt imize po lygon c ount of thederived surface while m aintaining t he o verall shap e.

By combining these techniques, surprisingly complex shapes can be created in just two or three

steps. Figure 9 shows the effects of extruding faces or points, beveling corners, co-planaring edges,

and other simple operations.

Figure 9 Starting with the c ube (top left), the effects of extruding fa ce s (top middle), extrudingpoints and then be veling c orners (top right), co -planaring ed ges (bottom left), bridging a holethrough the center (botto m midd le), and ve rtex-subd ividing a fa ce (b ottom right).


7/13



MODELINGFOR ANIMATION

Beyond the issue of how a control object reacts to smoothing, its necessary to attend to the derived

surfaces potential for animation. It needs to look right from any angle, and it needs to look right inall the various poses it might assume as its joints bend. To create these results, your modeling pro-

cess should take into account the three fundamental aspects of dynamic form: contour, silhouette,

and movement.

Silhouette is a models 2D outline from any given angle. A standard approach in traditional sculp-

ture is to look at your work from as many angles as possible. When the silhouette looks correct from

any angle, youre done. To avoid adversely affecting the silhouette in one perspective while youre

changing it in another, its best to constantly adjust the camera position while youre working. Many

artists end up working with the camera aligned on just the X, Y, and Z planes, especially when the

model becomes too complex to spin around freely.

Fortunately, a control object should never become so complex that you cant spin it around.

Contour is the 3D manifestation of silhouette. Among other things, it defines the way light fallsacross the model, making shadows a helpful tool in modeling. If you study the way a harsh light

falls across a characters head, one can see how the shadow forms a strong line running down the

support bone of the temple, across the wrinkles on the side of the eye, along the cheek to the jaw

line, and then down the neck (Fig 10a). Shadows can tell you at a glance how an object projects in

3D space. Watch them carefully as you work, and with practice, youll be able to recreate a 3D rep-

resentation fairly accurately from only a 2D image (Fig. 10b).

Figure 10 Harsh light fa lling a cross a t hree-dime nsional fo rm revea ls its co ntour. With prac tice , youllbe a ble to rec reate a 3D representation fa irly acc urately from only a 2D imag e.


8/13



The lighting within your modeling environment creates the shadows that help you evaluate silhou-

ette and contour. Therefore, its important to understand how the light behaves in a realtime model-

ing environment, viewed in perspective camera mode. For each pixel, the computer calculates the

way a ray of light bounces off the surface of the polygon and into the camera. The greater the angleof the bounce, the darker the pixel.

However, realtime hardware typically lights a scene based on the normals only (for speed reasons).

That is, real time renderers do shading, not shadows. If a side-lit character faces the camera and

opens its mouth, any polygons on the inside of his mouth whose normals face the light will also re-

flect it. This is something to watch out for when you examine contour and silhouette.

In most modeling packages, a light is "fixed" to the camera and moves with it automatically. In this

situation, no matter how you look at the model, it will be lit as though you were wearing a miners

hat with a headlamp (Fig. 11). Generally this is helpful, since the model will never pass into shad-

ow. But if the way a shadow falls across a model is integral to your modeling process, a different

lighting setup is in order.

Figure 11 In most mode ling p ac kages, a light is "fixed" to t he c ame ra and moves with it auto mat i-ca lly. This setup ob scures a m od els c onto ur.

Replace (or augment) the cameras default light with two infinite lights that also move with the cam-

era. Aim them directly at each other across the model diagonally, one pointing downward from the

upper right, one pointing upward from the lower left, and dim the lower light to 50% (Fig. 12).

Figure 12 To ma ke conto ur more visible as you mod el, replac e (or aug ment) the ca meras defa ultlight w ith two infinite lights that also m ove w ith the ca mera. Aim them d irec tly at ea ch o ther acrossthe mod el diagona lly, one pointing do wnwa rd from the upp er right, one pointing upw ard from thelowe r left, a nd d im the low er light t o 50%.


9/13



When used to light a model, this setup brings the contour into sharp relief (Fig. 13).

Figure 13 The mo dified lighting setup c reate s shad ow s tha t bring co ntour into sha rp relief.

Keep in mind that your object is going to move, twist, and bend, and that the static form youre cre-

ating will need to look good in any position. Test the models flexibility and evaluate the results in

terms of silhouette and contour. Theres no need to model characters in the familiar frog-dissection

pose. The traditional reason for this pose was to make it easier to skin the vertices to appropriate

bones and make it easier for the animator to define muscle behavior and the softwares skinning al-

gorithm is robust, you can work on the model in a more natural pose, which yields a better sense of

form, weight, and personality.

To see how the derived surface will react when the control object is posed in extreme positions, ro-

tate limbs straight up, straight out, and as far back as they can move. This will reveal problems such

as shearing or intersecting faces around the moving joint. The geometry of the control object must

accommodate a full range of motion. A common mistake is to add geometry around areas that arent

behaving correctly. Unfortunately, even with the added geometry, models like the one in Figure 14

break down rapidly when animated. A human shoulder joint is a floating joint that moves around

the center of the chest, not a ball-and-socket joint like the hip, so the extra geometry has been added

in the wrong place. More important, the extra geometry in Figure 14 makes no contribution to either

contour or silhouetteit's wasted data.

Figure 14 The ge om etry that ha s be en a dd ed t o this mo de ls joints will brea k when its anima ted .More important, the extra g eome try makes no c ontribution to co ntour or silhouette.


10/13



Figure 15 shows a better example of how to add geometry around a shoulder joint. By adding geom-

etry that flexes correctly during animation, you can make shoulders, armpits, and any other dynamic

forms a good deal more lifelike.

Figure 15 This mo de ls should er has be en b uilt correct ly, contributing t o the mo de ls c onto ur and sil-houette while a llowing for shoulder-type motion. By add ing ge ome try that flexes co rrec tly duringanimation, you ca n ma ke shoulders, armpits, and any othe r dynamic fo rms a g ood dea l more life-like.

EDGE LOOPS

Creating good contour edges on the control object has another beneficial effect. If you shape them

properly, theyll mimic the actual muscle structures that make up the body of the model. Selecting


11/13



vertices along these edge loops lets you manipulate the model as if you were moving the underlying

muscle. The key is to define an object topology made up of such edge loops (Fig. 16).

Figure 16 A control objects edges can mimic muscle structures.

Adjacent vertices form loops that allow you to manipulate a model as though you were moving the

underlying muscles. The key is to define an object topology made up of such edge loops.

The better these edge loops mimic the muscular structure of the body being animated, the more real-

istic the movement, contour, and silhouette of both the control object and derived surface will look.


12/13



Even trouble areas like the corners of the mouth can be posed into any position with excellent re-

sults (Fig. 17).

Figure 17 With we ll-designed edg e loop s, even trouble a reas such a s the c orners of the mouth ca nbe posed into a ny position without c reating silhouette, c ontour, or movement problems.


13/13



RAISINGTHE BAR

To produce high-quality work, content developers working in 3D face not only technological and

aesthetic challenges. They must cultivate an understanding of the traditional techniques of sculpt-ing, composition, and painting. Working with a computer doesnt exempt you from these consider-

ations. Rather, it gives you a more powerful and flexible set of tools with which to master age-old

artistic skills.

Because it both mimics and extends the traditional sculpting process, volume modeling provides a

direct path toward this goal. Working on the computer lets you reach an aesthetically pleasing

threshold so quickly that your only option for growth as an artist is to raise the bar. We hope the

techniques discussed in this article help you do it.

Bay Raitt is a digital sculptor and animator/director, currently working on the Lord of the Rings project with WETA. He has previously

been Engineering Product Specialist at Nichimen Graphics, lead modeler at San Francisco-based Protozoa, audio engineer, and comic

book color separator (Spawn, Akira, Pitt, and MAX) . He currently is involved in the quest for the perfect 3D pencil.

Greg Minter is the primary writer at Nichimen Graphics, responsible for all the documentation in Mirai.

digital sculpting

Documents